Blur

Blur is exactly what the name says: it blurs an image. Not a fancy anisotropic or depth-aware blur - a plain Gaussian or Box blur that runs on the CPU in a second. Which sounds boring until you remember how often a workflow needs a quick defocus. Softening a background for depth-of-field, killing compression noise before an upscale pass, smearing a glow layer, or making a face slightly less terrifyingly sharp. For those jobs a single node with a radius slider beats building a kernel by hand.

How it works

It's a thin wrapper over Pillow's ImageFilter. The IMAGE tensor converts to a PIL image, and depending on blur_type you get GaussianBlur or BoxBlur, both driven by radius. Gaussian is the smooth, natural-looking one - it spreads each pixel across a bell curve, so edges stay soft but recognizable. Box blur is a crude average over a square neighborhood: faster to compute, but you can see the blockiness at high radii. For most looks, stick with Gaussian. The only time Box earns its keep is when you're blurring huge images and the speed difference actually shows up.

Both types take a radius from 0 to 100 in 0.1 steps, default 5. Radius 0 passes the image through unchanged.

The inputs

• image - what you're blurring.
• blur_type - Gaussian or Box.
• radius - 0–100, default 5. That's the whole control surface.

Output is a single image tensor. It feeds forward into whatever you were doing - a Save node, a blend, or a mask pipeline.

One thing worth knowing: like every image node in this pack, it operates on the first frame only if you hand it a batched image (the tensor-to-PIL helper takes frame zero). Single images are the happy path here.

Installing it

Same as the rest of the pack - it ships in isekai-comfy-node, no separate install:

ComfyUI Manager: search "isekai" → install isekai-comfy-node → restart.

Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/isekai-sh/isekai-comfy-node
cd isekai-comfy-node
pip install -r requirements.txt

Then restart ComfyUI. It shows up under Isekai → Image/Effects. The only dependencies are Pillow and requests - torch and numpy are already in your ComfyUI environment. No models to download.

Common issues

There isn't much that can break here, which is the point of a node this simple. The errors you might actually see are the usual ComfyUI infrastructure ones: missing node after you forgot to restart, or the pack's dependencies not installed because you cloned without running pip install -r requirements.txt. If a workflow you downloaded is complaining about a missing IsekaiBlur, it's that - install the pack through Manager and let it handle the pip step.

Worth a mention: heavy blur is a known way to erase fine texture, so if you're blurring before an upscale, keep the radius small (2–4) or you'll end up smoothing away detail the upscaler can't bring back.